The emergence and evolution of micro-fractures in reservoirs establish the pore-fracture system as a primary storage space for oil and gas, as well as the main channels for hydrocarbon migration. A profound understanding of fracture morphology is vital for uncovering fracture propagation mechanisms, essential for assessing reservoir fracturability and enhancing productivity. High-resolution, non-destructive Computed Tomography (CT) scanning technology, coupled with three-dimensional (3D) fracture network reconstruction, is an effective tool for studying rock micro-features. This method allows for three-dimensional visualization and enables detailed, quantitative characterization of microstructural features. In this study, reservoir shale rock samples, characterized by well-developed stratification, were selected for analysis. Initially, a three-dimensional digital core model was established using CT scan image processing techniques. Image processing algorithms and deep learning techniques were then employed for precise fracture identification and extraction. Following a comparison of various deep learning architectures, the optimally tuned UNet2.5D architecture was found to be the most suitable for accurately identifying core fractures. To further quantitatively characterize shale fractures, the UNet2.5D model was utilized to identify shale fractures, and parameters including volume, surface area, theta, phi, sphericity, aspect ratio, and others were analyzed. This enabled a quantitative characterization of the fracture structure. The methodology proposed in this study for three-dimensional fracture identification and quantification in core samples contributes to the accurate assessment of reservoir micro-parameters and provides technical support for in-depth studies of fracture propagation mechanisms in complex environments.